CN108335716A - A kind of memory computational methods based on nonvolatile storage - Google Patents
A kind of memory computational methods based on nonvolatile storage Download PDFInfo
- Publication number
- CN108335716A CN108335716A CN201810077024.5A CN201810077024A CN108335716A CN 108335716 A CN108335716 A CN 108335716A CN 201810077024 A CN201810077024 A CN 201810077024A CN 108335716 A CN108335716 A CN 108335716A
- Authority
- CN
- China
- Prior art keywords
- memory
- storage unit
- storage
- nonvolatile
- denoted
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G11—INFORMATION STORAGE
- G11C—STATIC STORES
- G11C16/00—Erasable programmable read-only memories
- G11C16/02—Erasable programmable read-only memories electrically programmable
- G11C16/06—Auxiliary circuits, e.g. for writing into memory
- G11C16/10—Programming or data input circuits
-
- G—PHYSICS
- G11—INFORMATION STORAGE
- G11C—STATIC STORES
- G11C11/00—Digital stores characterised by the use of particular electric or magnetic storage elements; Storage elements therefor
- G11C11/02—Digital stores characterised by the use of particular electric or magnetic storage elements; Storage elements therefor using magnetic elements
- G11C11/16—Digital stores characterised by the use of particular electric or magnetic storage elements; Storage elements therefor using magnetic elements using elements in which the storage effect is based on magnetic spin effect
- G11C11/165—Auxiliary circuits
- G11C11/1675—Writing or programming circuits or methods
Abstract
A kind of memory computational methods based on nonvolatile storage of the present invention directly carry out logical calculated using non-volatile memory cells, as follows:For the typical 1T1R storage units of a nonvolatile memory, usually there are three signals for tool:Access control signal, is denoted as A, for accessing control to storage unit;Currently stored data, are denoted as Bi;Signal is written, is denoted as C;According to these three signals, the data of the next storage of storage unit are denoted as Bi+1, it is expressed asFinal logical calculated result, that is, Bi+1, it is stored directly in storage unit;Its logical calculated operates the normal consistent in reading and writing with memory.The present invention greatly alleviates storage wall and power consumption wall problem;The present invention directly carries out logical calculated using the storage unit of nonvolatile memory, and additional logic computing unit need not be integrated in memory chip, can substantially reduce the complexity and cost of memory computing chip.
Description
Technical field
The memory computational methods based on nonvolatile storage that the present invention relates to a kind of belonging to non-volatile memories and calculate neck
Domain.
Background technology
In traditional von Neumann counting system structure, memory is detached with processor, passes through data between the two
Bus is attached, and data processing needs carry out transmission back between memory and processor.But with big data application it is emerging
It rising, transmission and the processing of mass data make traditional von Neumann counting system structure face the double challenge of bandwidth and power consumption,
It is referred to as storage wall and power consumption wall problem.In order to solve the two problems, in recent years, memory calculates (Processing-In-
Memory, PIM) technology obtains the extensive research of people.The basic thought that convential memory calculates is by built-in in memory
Some logic computing units, to which some are simple, but data volume again prodigious logic computing function put it is complete in memory
At to reduce the data transmission between memory and processor.But logic computing unit and storage unit are integrated in one
Design complexities and manufacturing cost on block chip are all very high.Therefore, up to the present, memory computing technique does not obtain
Using.Recently, it with the continuous maturation of non-volatile memory technologies, is expected to provide new solution for the implementation of memory computing technique
Certainly scheme.
Non-volatile memory technologies, such as spin-transfer torque magnetic RAM (Spin Transfer Torque
Magnetic Random Access Memory, STT-MRAM), phase-change random access memory (Phase Change Random
Access Memory, PCRAM) and resistor type random access memory (Resistive Random Access Memory, RRAM) etc.
It continues to bring out, has gradually started large-scale production and application.Due to many good characteristics of nonvolatile memory, such as data
Power down is not lost, low-power consumption, at high speed etc., there is very important meaning in Future Data storage and logical calculated field
Justice.The typical memory cell of these nonvolatile memories is by a transistor (being denoted as T) and a nonvolatile semiconductor memory member
(being considered as a variable resistance, be denoted as R) forms, referred to as 1T1R memory cell structures, as shown in Fig. 1, wherein resistive device
For storing binary data information, transistor is used to access control to storage unit part.Up to the present, non-volatile
Memory technology is mainly used for storing data.The it is proposed of the invention is a kind of directly to be carried out using non-volatile memory cells
The method of logical calculated, need not integrate additional logic computing unit in memory, and storage unit itself can store
Data can also carry out logical calculated, to realize that memory truly calculates.
Invention content
One, goal of the invention:
For under the big data application scenarios mentioned in above-mentioned background, what traditional von Neumann counting system structure faced deposits
Wall and power consumption wall problem are stored up, the present invention discloses a kind of memory computational methods based on nonvolatile storage.It is calculated with convential memory
Technology is different, and the present invention directly carries out logical calculated using non-volatile memory cells.It in this way need not be in memory
Additional logic computing unit is integrated in chip, greatly reduces the complexity and cost of memory computing chip.
Two, technical solution:
The technical scheme is that:A kind of memory computational methods based on nonvolatile storage directly utilize non-easy
The property lost Memory Storage Unit carries out logical calculated.Detailed process is as follows:For the typical 1T1R of a nonvolatile memory
There are three signals for storage unit, usually tool:(1) access control signal (being denoted as A), for accessing control to storage unit;
(2) currently stored data (are denoted as Bi);(3) write-in signal (being denoted as C).According to these three signals, storage unit is next to deposit
The data of storage (are denoted as Bi+1) can be expressed asWherein, write-in signal C is considered as logic function selection
Signal determines the power function of logical calculated, for example, when C is respectively equal to " 0 ", " 1 ", or" when, one under storage unit
The data B of a storagei+1Respectively equal to("AND" logic), A+Bi("or" logic) or(exclusive logic).Finally
Result (the i.e. B of logical calculatedi+1) be stored directly in storage unit.Its logical calculated operates the normal read-write with memory and grasps
Make consistent.Method proposed by the present invention can be applied equally to other Nonvolatile memery units with similar signal feature
Structure.
Can the access control signal A be accessed for controlling storage unit, unlimited specific electrical form, can be with
It is electric current, other physical forms such as voltage or magnetic field.
The currently stored data Bi, physically it is expressed as the resistance of nonvolatile semiconductor memory member in current memory cell
State can be respectively high-impedance state or low resistance state two states.
Said write signal C, for carrying out data write-in to storage unit, unlimited specific electrical form can be electricity
Stream, other physical forms such as voltage or magnetic field.
Three, advantage and effect:
The memory computational methods based on nonvolatile storage that the present invention provides a kind of, need not be in memory and processor
Between moving data back and forth, therefore can greatly alleviate storage wall and power consumption wall problem.Meanwhile the present invention is directly using non-easy
The storage unit of the property lost memory carries out logical calculated, and additional logic computing unit need not be integrated in memory chip,
Therefore, the complexity and cost of memory computing chip can be greatly reduced.
Description of the drawings
Fig. 1 is the typical 1T1R memory cell structures schematic diagram of nonvolatile memory of the present invention.
Fig. 2 is a kind of logic that the memory computational methods kind storage unit based on nonvolatile memory is related to letter of the present invention
Number.
Fig. 3 (a), (b), (c) are a kind of corresponding state of memory computational methods based on nonvolatile memory of the present invention
Transfer figure, truth table and expression formula.
Fig. 4 (a), (b), (c) are a kind of memory computational methods based on spin-transfer torque magnetic RAM of the present invention
Specific embodiment.
Parameter definition in figure is:
BL:It indicates bit line, is the abbreviation of Bit-Line;
WL:It indicates wordline, is the abbreviation of Word-Line;
SL:It indicates source electrode line, is the abbreviation of Source-Line;
RL:Nonvolatile semiconductor memory member is in the resistance value of low-resistance state of value;
RH:Nonvolatile semiconductor memory member is in the resistance value of high resistant state of value;
Rdata:The resistance value of nonvolatile semiconductor memory member, there is RHAnd RLTwo kinds of possibility;
A:Can access control signal be accessed for controlling storage unit;
C:Signal is written, for carrying out data write-in to storage unit;
Bi:The currently stored data of storage unit;
STT-MRAM:Spin-transfer torque magnetic RAM
MTJ:Magnetic tunnel-junction, the basic unit of spin-transfer torque magnetic RAM have high-impedance state and low resistance state two
State.
Specific implementation mode
The present invention provides a kind of memory computational methods based on nonvolatile memory.With reference to attached drawing, this is further illustrated
The substantive distinguishing features of invention.Attached drawing is schematic diagram, and the parameters such as each size of devices being directed to not are actual size.
It is disclosed that detailed exemplary embodiment, specific CONSTRUCTED SPECIFICATION and function detail are only the specific realities of description
Apply the purpose of example, therefore, can by it is many it is selectable in the form of implement the present invention, and the present invention is not construed as only
It is confined to the example embodiment herein proposed, but all changes fallen within the scope of the present invention, equivalent and can should be covered
Alternative.In addition, it will not be described in detail or will omit the well-known element of the present invention, device and sub-circuit, so as not to it is mixed
Confuse the correlative detail of the embodiment of the present invention.
Fig. 1 is the typical 1T1R memory cell structures schematic diagram of nonvolatile memory of the present invention.
The typical memory cell of nonvolatile memory is by a transistor (being denoted as T) and a nonvolatile semiconductor memory member
(being considered as a variable resistance, be denoted as R) forms, referred to as 1T1R memory cell structures.Wherein nonvolatile semiconductor memory member is used
In storage data information, resistance value can be there are two types of state, and one kind is high value state (RH), one kind is low resistance state (RL), respectively
Represent data bit " 0 " and " 1 " or on the contrary.Transistor is used to access control to storage unit, and grid connects wordline,
Drain electrode is followed by bit line via nonvolatile semiconductor memory member, and source electrode connects source electrode line, and source electrode line is typically grounded.By the electricity for controlling wordline
Pressure can controlling transistor opening and closing, whether selection to control storage unit.More specifically, when wordline is high level,
Transistor is in the conduction state, and storage unit may have access to, and operation can be written and read to it;And when wordline is low level, crystal
Pipe is in nonconducting state, and storage unit is inaccessible.
2, attached drawing 3 and attached drawing 4 below in conjunction with the accompanying drawings, the specific implementation mode that the present invention will be described in detail.
A kind of memory computational methods based on nonvolatile storage of the present invention, directly utilize nonvolatile memory to store
Unit carries out logical calculated.As shown in Fig. 2, for a nonvolatile memory typical 1T1R storage units (such as Fig. 1 institutes
Show), usually there are three signals for tool:(1) access control signal (being denoted as A), for accessing control to storage unit;(2) when
The data of preceding storage (are denoted as Bi);(3) write-in signal (being denoted as C).According to these three signals, the number of the next storage of storage unit
According to (being denoted as Bi+1) can be expressed asFig. 3 (a), (b), (c) show corresponding state transition diagram, true value
Table and expression formula.As can be seen that write-in signal C is considered as logic function selection signal, the function of logical calculated is determined
Function, for example, when C is equal to " 0 ", " 1 ", orWhen, the data B of the next storage of storage uniti+1Respectively equal to
("AND" logic), A+Bi("or" logic) or(exclusive logic).Result (the i.e. B of final logical calculatedi+1) directly
It is stored in storage unit.Its logical calculated operates the normal consistent in reading and writing with memory.Method proposed by the present invention is same
Sample can be applied to other non-volatile memory cell structures with similar signal feature.
Can the access control signal A be accessed for controlling storage unit, unlimited specific electrical form, can be with
It is electric current, other physical forms such as voltage or magnetic field.
The currently stored data Bi, physically it is expressed as the resistance of nonvolatile semiconductor memory member in current memory cell
State can be respectively high-impedance state or low resistance state two states.
Said write signal C, for carrying out data write-in to storage unit, unlimited specific electrical form can be electricity
Stream, other physical forms such as voltage or magnetic field.
Embodiment one:By taking spin-transfer torque magnetic RAM as an example, the specific embodiment that the present invention will be described in detail.
As shown in figure 4, the typical 1T1R storage units of spin-transfer torque magnetic RAM are by a transistor and one
Information, transistor are used to access to storage unit for storing data for a magnetic tunnel-junction (MTJ) composition, wherein magnetic tunnel-junction
Control.In spin-transfer torque magnetic RAM, access control signal A is the grid of transistor in 1T1R memory cell structures
Pole tension, high level are " 1 ", and low level is " 0 ".The write current direction that signal C is 1T1R memory cell structures, " 0 " is written
Representative writes " 0 " to MTJ, and " 1 " is represented to MTJ one writings.Currently stored data BiIndicate the resistance states of current MTJ, RHIt indicates
High-impedance state represents data " 0 ", RLIt indicates low resistance state, represents data " 1 ", or vice versa.When carrying out memory calculating, such as Fig. 4
(a) shown in, if C=0,It executesWith BiBetween "AND" logical calculated;As shown in Fig. 4 (b), if C
=1, then Bi+1=A+Bi, execute A and BiBetween "or" logical calculated;As shown in Fig. 4 (c), ifThenExecute A and BiBetween exclusive logic calculate.
Claims (5)
1. a kind of memory computational methods based on nonvolatile storage, it is characterised in that:Directly deposited using nonvolatile memory
Storage unit carries out logical calculated, and detailed process is as follows:
For the typical 1T1R storage units of a nonvolatile memory, usually there are three signals for tool:Access control signal, note
For A, for accessing control to storage unit;Currently stored data, are denoted as Bi;Signal is written, is denoted as C;
According to these three signals, the data of the next storage of storage unit are denoted as Bi+1, can be expressed as
Final logical calculated as a result, i.e. Bi+1, it is stored directly in storage unit;Its logical calculated operates the normal reading with memory
Write operation is consistent.
2. a kind of memory computational methods based on nonvolatile storage according to claim 1, it is characterised in that:Described
Write-in signal C is considered as logic function selection signal, determines the power function of logical calculated, when C is respectively equal to " 0 ",
" 1 ", or" when, the data B of the next storage of storage uniti+1Respectively equal toThat is "AND" logic, A+BiI.e. "or" is patrolled
Volume, orThat is exclusive logic.
3. a kind of memory computational methods based on nonvolatile storage according to claim 1, it is characterised in that:Described
Can access control signal A be accessed for controlling storage unit, unlimited specific electrical form, can be electric current, voltage,
Or magnetic field.
4. a kind of memory computational methods based on nonvolatile storage according to claim 1, it is characterised in that:Described
Currently stored data Bi, the resistance states of nonvolatile semiconductor memory member in current memory cell are physically expressed as, respectively
High-impedance state or low resistance state two states.
5. a kind of memory computational methods based on nonvolatile storage according to claim 1 or 2, it is characterised in that:Institute
The write-in signal C stated, for carrying out data write-in to storage unit, unlimited specific electrical form can be electric current, voltage,
Or magnetic field.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810077024.5A CN108335716B (en) | 2018-01-26 | 2018-01-26 | Memory computing method based on nonvolatile memory |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810077024.5A CN108335716B (en) | 2018-01-26 | 2018-01-26 | Memory computing method based on nonvolatile memory |
Publications (2)
Publication Number | Publication Date |
---|---|
CN108335716A true CN108335716A (en) | 2018-07-27 |
CN108335716B CN108335716B (en) | 2020-12-15 |
Family
ID=62926471
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201810077024.5A Active CN108335716B (en) | 2018-01-26 | 2018-01-26 | Memory computing method based on nonvolatile memory |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN108335716B (en) |
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110096308A (en) * | 2019-04-24 | 2019-08-06 | 北京探境科技有限公司 | A kind of parallel memorizing arithmetic unit and its method |
CN110459254A (en) * | 2019-08-01 | 2019-11-15 | 北京航空航天大学 | Spin electric device and deposit interior logic calculation device |
CN111045727A (en) * | 2018-10-14 | 2020-04-21 | 天津大学青岛海洋技术研究院 | Processing unit array based on nonvolatile memory calculation and calculation method thereof |
US11443185B2 (en) | 2018-10-11 | 2022-09-13 | Powerchip Semiconductor Manufacturing Corporation | Memory chip capable of performing artificial intelligence operation and method thereof |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103716038A (en) * | 2013-12-25 | 2014-04-09 | 华中科技大学 | Nonvolatile logic gate circuit based on phase change memories |
CN103730149A (en) * | 2014-01-20 | 2014-04-16 | 中国科学院电子学研究所 | Read write control circuit of dual-ported memory |
CN107017020A (en) * | 2017-02-22 | 2017-08-04 | 北京航空航天大学 | A kind of data of nonvolatile storage read method |
CN108182959A (en) * | 2018-01-22 | 2018-06-19 | 中国科学院微电子研究所 | The method that logical calculated is realized based on resistive device cross array structure |
CN109660250A (en) * | 2018-12-03 | 2019-04-19 | 北京大学深圳研究生院 | A kind of polymorphic door based on resistance-variable storing device |
-
2018
- 2018-01-26 CN CN201810077024.5A patent/CN108335716B/en active Active
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103716038A (en) * | 2013-12-25 | 2014-04-09 | 华中科技大学 | Nonvolatile logic gate circuit based on phase change memories |
CN103730149A (en) * | 2014-01-20 | 2014-04-16 | 中国科学院电子学研究所 | Read write control circuit of dual-ported memory |
CN107017020A (en) * | 2017-02-22 | 2017-08-04 | 北京航空航天大学 | A kind of data of nonvolatile storage read method |
CN108182959A (en) * | 2018-01-22 | 2018-06-19 | 中国科学院微电子研究所 | The method that logical calculated is realized based on resistive device cross array structure |
CN109660250A (en) * | 2018-12-03 | 2019-04-19 | 北京大学深圳研究生院 | A kind of polymorphic door based on resistance-variable storing device |
Non-Patent Citations (2)
Title |
---|
H. ZHANG, W. KANG, L. WANG, K. L. WANG 和 W. ZHAO: "Stateful Reconfigurable Logic via a Single-Voltage-Gated Spin Hall-Effect Driven Magnetic Tunnel Junction in a Spintronic Memory", 《IEEE TRANSACTIONS ON ELECTRON DEVICES》 * |
KANG,W;ZHANG,H;OUYANG,P;ZHANG YG;ZHAO,WS: "Programmable Stateful In-Memory Computing Paradigm via a Single Resistive Device", 《2017 IEEE 35TH INTERNATIONAL CONFERENCE ON COMPUTER DESIGN (ICCD)》 * |
Cited By (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US11443185B2 (en) | 2018-10-11 | 2022-09-13 | Powerchip Semiconductor Manufacturing Corporation | Memory chip capable of performing artificial intelligence operation and method thereof |
CN111045727A (en) * | 2018-10-14 | 2020-04-21 | 天津大学青岛海洋技术研究院 | Processing unit array based on nonvolatile memory calculation and calculation method thereof |
CN111045727B (en) * | 2018-10-14 | 2023-09-05 | 天津大学青岛海洋技术研究院 | Processing unit array based on nonvolatile memory calculation and calculation method thereof |
CN110096308A (en) * | 2019-04-24 | 2019-08-06 | 北京探境科技有限公司 | A kind of parallel memorizing arithmetic unit and its method |
CN110459254A (en) * | 2019-08-01 | 2019-11-15 | 北京航空航天大学 | Spin electric device and deposit interior logic calculation device |
CN110459254B (en) * | 2019-08-01 | 2021-06-01 | 北京航空航天大学 | Spintronic device and memory logic computing device |
Also Published As
Publication number | Publication date |
---|---|
CN108335716B (en) | 2020-12-15 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
US10777275B2 (en) | Reset refresh techniques for self-selecting memory | |
CN108335716A (en) | A kind of memory computational methods based on nonvolatile storage | |
US11087840B2 (en) | Method of operating resistive memory device to increase read margin | |
US8004872B2 (en) | Floating source line architecture for non-volatile memory | |
US10269423B2 (en) | Access methods of memory device using relative addressing | |
CN109545258A (en) | Resistive memory device and its operating method including reference unit | |
Chen et al. | Recent technology advances of emerging memories | |
US20190164581A1 (en) | Sense amplifier with comparison node biasing for non-volatile memory | |
EP3417456A1 (en) | Dual demarcation voltage sensing before writes | |
CN113785290A (en) | Memory-based vector-matrix multiplication | |
US8203894B2 (en) | Current cancellation for non-volatile memory | |
US11443801B2 (en) | Semiconductor memory apparatus for preventing disturbance | |
Chang et al. | Read circuits for resistive memory (ReRAM) and memristor-based nonvolatile Logics | |
US11869595B2 (en) | Memory device programming technique using fewer latches | |
US20220180930A1 (en) | Binary to ternary convertor for multilevel memory | |
CN111916127B (en) | Resistive random access memory unit and resistive random access memory | |
US10622065B2 (en) | Dedicated commands for memory operations | |
KR20130021197A (en) | Non-volatile device of semiconductor memory | |
CN102157196B (en) | ITIR resistive random access memory based on self-reference inverter and reading and writing method thereof | |
CN109461466A (en) | Resistive memory devices and its operating method with reduced chip size | |
US20230307043A1 (en) | Technologies for current biasing for memory cells | |
US20230093729A1 (en) | Technologies for on-memory die voltage regulator | |
CN102890963A (en) | Non-volatile random access memory and operation method thereof | |
WO2023029539A1 (en) | Non-volatile memory, and writing method and reading method therefor | |
KR20200090289A (en) | Memory device and operation method thereof |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |